Reframe - AI Agents on a non-code database platform

Introducing Reframe AI, an AI database with Agents (virtual assistants) that do work for you. Reframe combines a low-code interface with AI agents to automate data workflows end-to-end, accelerate queries, eliminate bottlenecks, enrich data, run analytics, generate insights, and optimize data management.

Reframe enables you to create executable workflows that link Large Language Models (LLMs), Prompts and Python functions together in a directed acyclic graph. With Reframe, you can create complex workflows that operate on data tables - thereby taking advantage of the similarities and co-dependencies amongst data.

Reframe strives to be

  • :goggles: Transparent - through logging, and metrics that create visibility into the inner operations.

  • :person_cartwheeling:t5: Flexible - AI Agents and tools are independent of each other, allowing you to create workflows easily.

  • :jigsaw: Composable. AI agents and tools are simply executable python functions and classes with a well defined interface. You can easily construct sophisticated agents from basic building blocks. These building blocks can be obtained from our ecosystem or you can develop custom ones that are specific to your organization.

  • :skateboard: Incrementally adoptable - By using existing technologies such as Docker, Kubernetes and Celery Reframe enables you to seamlessly adopt it with your organization. From simple ZeroShot agents to sophisticated multi-step AI agents each component can be integrated into your existing workflows.

  • :hammer: Reusable - once a tool is running, it can be utilized by various agents, thereby reducing operational overhead, increasing throughput and making tools easy to reason about.

  • :racing_car: Fast by taking advantage of data parallelism and prompt sequencing in a manner increases efficiency and reduces the overall number of expensive API calls made to LLM endpoints.

  • :stadium: Rich ecosystem that enables your to pick and choose which tools and agents to deploy. Through contributions from open source developers, we are making great progress to develop a robust ecosystem of tools so that you always have a tool for the job.

What are LLM AI Agents?

Simply put, AI Agents are a means for LLMs to interact with the outside world via tools controlled by a module that has control logic and memory. An example of a tools is Google Search or LinkedIn API. While immensely useful, LLMs on their own lack many of the features that a normal user would interact with in their day to day life. Simply prompting

What is Reframe?

At Reframe we believe that the true power of LLMs lies in agentic behavior. By engineering a system that draws on LLMs’ emergent abilities and providing an ecosystem that supports environmental interactions for tabular data, we can fully realize the full potential of models like GPT-4.

Youtube Preview

Hey there, Welcome to OpenAI developer forum.

Thanks for sharing Reframe. It’s cool to see a Python framework like this. I’m curious about its real-world applications. Got any interesting use cases to share? Also, if there’s more info or examples, I’d love to check them out.

Looking forward to learning more,


Hi @wwwarmy38. Here are some real world examples of how ReframeAI could be used. They key is that you instruct an AI agent (Virtual assistant) to run a particular task for you - such as google search. The results of this agent are then processed by a LLM extracting the information that you need.

Sales and Prospecting for B2B startups:

  • Automate lead research and qualification at scale - Instruct ReframeAI agents to compile prospect details, analyze websites, and score leads to identify qualified accounts worth prioritizing. This replaces manual research and vetting.

  • Provides real-time alerts and monitoring - Reframe’AI agents continuously monitor news, events, and triggers relevant to prospects so sales can capitalize on timely outreach opportunities. This contextual awareness facilitates breaking through the noise.

  • Delivers actionable intelligence on prospects - ReframeAI enriches CRM data with supporting details like tech stack, initiatives, roles, and content extracts so sales has credible insights for meaningful prospect conversations. This eliminates cold outreach.

Commercial Real Estate:

  • Google search agents aggregate statistics, business listings, and other data points to build a complete view of each company’s growth trajectory. The agents also screen for negative signals like layoffs to filter out high-risk leads. This level of automated research and analysis is impossible for agents to do manually across a prospect list.
  • Browser agents rapidly research companies by analyzing press releases, news articles, competitor sites, and other web pages to uncover signals like new funding, mergers, growth plans, and personnel changes. This automates hours of manual online research.
  • LinkedIn agents identify key decision maker profiles and analyze recent hiring trends to detect location-specific roles like Office Managers that indicate imminent office growth needs. This provides insights that would take ages to manually uncover.

Wealth Management:
ReframeAI agents simplify lead research and prospecting for wealth managers:

  • Property analysis agents can rapidly uncover high net worth HNW prospects by cross-referencing luxury home purchases, voter registration data, LinkedIn, and mortgage records to identify unemployed women with wealthy spouses. This level of automated data synthesis would take a human days.

  • LinkedIn agents can identify up-and-coming professionals poised for higher earnings based on titles, employers, promotions, education, connections, and other indicators. The AI analyzes these signals at scale to surface emerging prospects.

  • News monitoring agents track funding rounds, exits, IPOs, and local expansions to alert wealth managers about liquidity events and income jumps. This real-time awareness allows outreach while prospects are still forming advisor relationships.

ReframeAI agents simplify ecommerce product management in the following ways:

  • Web crawling agents can scrape pricing, reviews, specs, and other product data from webpages at scale. This automates tedious manual research and consolidation of insights from multiple sites.

  • Reporting agents generate analytics on sales trends, price optimization, customer sentiment, vendor performance, and stock levels over custom time periods. Humans would spend hours compiling these data views.

  • Spreadsheet analysis agents mine product data like inventory, costs, margins, and attributes to uncover optimization opportunities. This level of granular analysis isn’t feasible manually across large datasets.

ReframeAI agents simplify recruiting workflows in the following ways

  • Sourcing agents rapidly identify and extract info from candidate profiles on sites like LinkedIn and the web based on parameters like skills, titles, and keywords. This automates manually searching for and documenting prospect details.

  • Research agents uncover insights on companies, roles, layoffs, finances, culture, and events during a candidate’s tenure. This level of background investigation would take recruiters hours per prospect.

  • Communication agents handle scheduling, screening, and note taking tasks. For example, chatbots interview and evaluate candidates to filter prospects before human interaction. This removes low value interactions.

Startup Investors and VCs:

  • Monitoring agents track emerging sources like Show HN, accelerators, subreddits, and Github to uncover startups just getting started before wider visibility. This early signal detection would require extensive manual effort.

  • Research agents compile candid founder reviews from LinkedIn connections, school alumni, and platforms like Crunchbase. The qualitative insights would take investor associates hours of outreach to gather.

  • Traction agents analyze Twitter, Reddit, forums, downloads, press, and other channels to identify which new brands are gaining grassroots enthusiast momentum. Manually monitoring this buzz is not scalable.

We are still working on releasing documentation for the python framework and I’ll ping you as soon as it’s available.